Perplexity Announces Portable Computer for Windows: Local AI and a Crusoe AI Partnership

5 min read

Originally published in Turkish. Read the original →

Perplexity Announces Portable Computer for Windows: Local AI and a Crusoe AI Partnership - cover image
Summary

Perplexity AI is accelerating its shift from a search engine to an autonomous agent ecosystem integrated into the operating system. Details on local PPLX 27B inference with Portable Computer for Windows, the Nvidia GB300-based Crusoe data center partnership and the retiring Sonar API.

Perplexity AI, which shook up the industry by turning the concept of a search engine into an AI-powered answer and research engine, is going through a strategic evolution as of September 2026. A journey that began with cloud-based information search models now extends to local AI agents running directly on users' hardware and to massive data center infrastructure. The "Portable Computer" tool the company officially announced for the Windows operating system, and the multi-year partnership it signed with next-generation hardware provider Crusoe AI, show that Perplexity is building not just a Q&A platform but a comprehensive enterprise AI ecosystem.

Portable Computer for Windows: Cloud-Independent Local Intelligence

The Perplexity Computer concept, previously offered to macOS users, has stepped into the Windows ecosystem under the name "Portable Computer." This new release lets AI models run entirely locally (on-device) on modern Windows PCs with an Nvidia GeForce RTX or RTX PRO graphics card with at least 24 GB of VRAM.

At the heart of the system is the PPLX 27B open-weight language model, specially optimized by Perplexity. Without sending any data to cloud servers or consuming cloud credits, users can analyze sensitive corporate documents, automate long-running research processes and run autonomous agents on their personal file systems. Especially for finance, legal and healthcare, sectors sensitive to data privacy and strict regulation, this local mode offers a secure alternative that removes cloud dependency. Professionals working in the field or in environments with limited internet access, such as on a plane, can keep their full research and analysis capability on their devices without interruption.

Deep Integration with Hardware Manufacturers

Perplexity's vision of computer-use agents is not limited to the software layer. The company is entering direct partnerships with leading hardware manufacturers to make AI a factory standard on devices. The most concrete example of this was seen in HP's next-generation mobile workstation, the ZBook Ultra G3a. The device shipped with the Perplexity Computer infrastructure preinstalled at the factory.

This integration lets professional users manage multi-step data analysis, code compilation, system automation and research processes right out of the box with an AI agent running directly at the operating system level. The workstation's powerful NPU and GPU units ensure that the AI agent running in the background consumes minimal system resources.

The Crusoe AI Partnership and a Next-Generation Infrastructure Move

In parallel with these steps on the local inference side, Perplexity signed a major multi-year agreement with infrastructure provider Crusoe to train frontier models in the cloud and handle the inference load in production. Under this partnership, announced on September 15, 2026, Perplexity will use Crusoe's green-energy-focused data centers powered by Nvidia GB300 NVL72 supercomputer clusters.

This high-density compute power will let Perplexity train its next-generation reasoning engines much faster while enabling the growing volume of user queries worldwide to be answered within milliseconds. The company thus gets past the hardware bottlenecks created by traditional hyperscale cloud providers and gains independent compute capacity.

Scientific Research and the Q2D-Web Retrieval Architecture

Perplexity's research lab does not just build products; it also produces work that sets academic standards in retrieval architectures. The company's latest technical papers reveal the details of the Q2D-Web retrieval architecture, which can filter the most relevant content from billions of pages on the web in its first stage. Going far beyond the keyword matching used by traditional search engines, the system maps semantic context and logical causality within milliseconds. As a result, all Perplexity agents, whether running locally or in the cloud, are fed with reliable sources that minimize the risk of hallucination.

The Agent API Era for Developers

The platform's technical backbone is also undergoing a significant modernization. Perplexity announced that it will fully retire the aging Sonar chat API endpoint as of September 27, 2026. The developer community is being directed to the next-generation Agent API architecture, which includes multi-step planning, function calling (tool use) and dynamic web browsing capabilities.

The new Agent API makes it much easier for developers to embed autonomous research agents in their own applications. Because the model cross-verifies multiple web sources simultaneously in the background while generating a response and delivers structured data with direct citations, it raises the reliability standard for API-based AI products.

Financial Growth and Market Dominance

According to the latest figures, Perplexity serves more than 30 million monthly active users and over 780 million search queries per month. Having passed the $100 million mark in annual recurring revenue (ARR), the company has reached an $18 billion valuation, cementing its leadership in the AI-powered search market.

In Conclusion

The Portable Computer tool Perplexity announced for Windows, its infrastructure investments crowned by the Crusoe partnership and its Agent API-focused developer transformation clearly show that the company is not only shaking the throne of traditional search engines but also shaping the future of autonomous agents integrated into operating systems. This hybrid model, which fuses local privacy with the frontier intelligence of the cloud, will continue to steer the course of generative AI in 2026 and beyond.